Offline Agent Using Reinforcement Learning to Speedup Trajectory Planning for Autonomous Vehicles

2. The method of claim 1, wherein the plurality of discretized control action options are generated based on a vehicle dynamic model for autonomous driving.

3. The method of claim 1, wherein the plurality of discretized trajectory state options are generated by discretizing a region of interest for the driving scenario in view of a final destination trajectory state.

4. The method of claim 1, wherein the judgment score includes scores representing whether the trajectory ends at a planned destination state, the trajectory is smooth, and the trajectory avoids one or more obstacles of an environment model.

5. The method of claim 1, wherein the driving scenario includes one or more regions of interest (ROIs).

6. The method of claim 1, wherein the RL agent includes an actor neural network and a critic neural network, and wherein the actor and critic neural networks are deep neural networks.

7. The method of claim 6, wherein the actor neural network includes a convolutional neural network.

9. The non-transitory machine-readable medium of claim 8, wherein the plurality of discretized control action options are generated based on a vehicle dynamic model for autonomous driving.

10. The non-transitory machine-readable medium of claim 8, wherein the plurality of discretized trajectory state options are generated by discretizing a region of interest for the driving scenario in view of a final destination trajectory state.

11. The non-transitory machine-readable medium of claim 8, wherein the judgment score includes scores representing whether the trajectory ends at a planned destination state, the trajectory is smooth, and the trajectory avoids one or more obstacles of an environment model.

12. The non-transitory machine-readable medium of claim 8, wherein the driving scenario includes one or more regions of interest (ROIs).

13. The non-transitory machine-readable medium of claim 8, wherein the RL agent includes an actor neural network and a critic neural network, and wherein the actor and critic neural networks are deep neural networks.

14. The non-transitory machine-readable medium of claim 13, wherein the actor neural network includes a convolutional neural network.

16. The data processing system of claim 15, wherein the plurality of discretized control action options are generated based on a vehicle dynamic model for autonomous driving.

17. The data processing system of claim 15, wherein the plurality of discretized trajectory state options are generated by discretizing a region of interest for the driving scenario in view of a final destination trajectory state.

18. The data processing system of claim 15, wherein the judgment score includes scores representing whether the trajectory ends at a planned destination state, the trajectory is smooth, and the trajectory avoids one or more obstacles of an environment model.

19. The data processing system of claim 15, wherein the driving scenario includes one or more regions of interest (ROIs).

20. The data processing system of claim 15, wherein the RL agent includes an actor neural network and a critic neural network, and wherein the actor and critic neural networks are deep neural networks.

21. The data processing system of claim 20, wherein the actor neural network includes a convolutional neural network.